Key takeaways
- Redshift reports size in SVV_TABLE_INFO, in 1 MB blocks, for the database you are connected to; add up the size column, divide by 1024 and report GB.
- The size is compressed and includes rows marked for deletion, so it will not match an uncompressed export; use estimated_visible_rows for live row counts.
- Each table is exported with UNLOAD to Amazon S3, using an IAM role on the cluster; Parquet is the usual format, and a verbose manifest lists every file.
- You don't need to export anything to apply: have your size in GB, years of history and customer or user count ready, and the data vendor anonymizes the records before any AI lab sees them.
01
What is Amazon Redshift data, and why do AI labs and data vendors buy it?
The short answer
Amazon Redshift data is the set of tables a company keeps in Redshift, the data warehouse that Amazon Web Services (AWS) runs: orders, billing, product usage and marketing records, cleaned and joined into tables. AI labs (companies that build and train AI models) and data vendors (companies that anonymize business records and package them for those labs) buy a copy because years of joined, row-level history show how customer behavior connects to revenue and retention.
What a Redshift warehouse records
A data warehouse is a database built to analyze large volumes of records copied in from a company's other systems. Amazon Redshift is the warehouse offered by AWS, Amazon's cloud computing service. A Redshift cluster, the group of computers that stores and queries the data, can hold several databases, and each database holds tables.
Warehouse tables are usually modeled as fact tables and dimension tables. A fact table has one row for each thing that happened, such as an order or a payment, and a dimension table lists the things those rows refer to, such as customers or products. Keys link the two, so each order can be traced to its customer and its product.
Why modeled history teaches an AI model something
An AI model learns by studying a very large number of examples and finding the patterns in them, and those examples are called training data. A warehouse holds years of orders, payments and usage with exact dates, so a model can see which customers kept buying, which stopped and what happened just before. That is real ground truth about business outcomes, and public text almost never contains it.
AI labs now want to train AI agents, software that carries out a job step by step, and an agent learns a job from records of that job being done. A data vendor can turn anonymized fact and dimension tables into worked examples that an agent studies, or into a practice copy of a business process in which the agent is scored on each attempt. The full mechanism is in AI training data: why AI companies want your company's data.
How Amazon Redshift fits into a dataset
A Redshift warehouse is usually one of several sources in a company's dataset, next to email, documents and tickets. Companies are paid $100K to $2M or more per dataset, according to micro1, the data vendor we typically refer companies to. The price covers the whole dataset, not each source on its own, and what your data is worth explains what moves it.
02
Where does Amazon Redshift show how much data you have?
The short answer
Amazon Redshift reports table sizes in the SVV_TABLE_INFO system view, counted in blocks of 1 MB. A superuser (a database administrator account with every privilege) can query it and can grant read access on it to other users. Add up the size column, divide by 1024 and report the total in gigabytes (GB).
How to read your storage total
The steps below are run by a developer or database administrator. They follow the SVV_TABLE_INFO reference.
- Connect as a superuser, in Query Editor v2 (the browser-based SQL tool in the AWS console) or in any SQL client. To let a colleague run the query as well, a superuser runs
GRANT SELECT ON svv_table_info TO username;. No special plan is needed. - Inside each database, run this statement. It groups the tables by schema and shows their size in megabytes, their live rows and the date the first table was created:
SELECT "schema", SUM(size) AS size_mb, SUM(estimated_visible_rows) AS visible_rows, MIN(create_time) AS first_table_created FROM svv_table_info GROUP BY "schema" ORDER BY size_mb DESC; - Divide the total of
size_mbby 1024 to get GB. The view covers only the database you are connected to, so connect to each database in turn and add the results. It also skips tables that are empty. - For a view of the whole cluster, look at the
PercentageDiskSpaceUsedmetric in CloudWatch, the AWS monitoring service. It gives the percentage of disk in use, which makes a useful cross-check but is not a size. - Find how far back the data goes by asking your largest fact tables for their oldest date.
What the SVV_TABLE_INFO columns count
| Column | What it counts | Use it? |
|---|---|---|
size | The space the table occupies, in 1 MB data blocks, after column compression. It includes rows marked for deletion that the cleanup process called vacuum has not yet removed. | Yes. Add it up and report it in GB. |
tbl_rows | Every row, including rows marked for deletion. | No, for live row counts. |
estimated_visible_rows | The estimated number of live rows, leaving out those marked for deletion. | Yes, for row counts. |
create_time | When the table was created. | Only as a rough guide to the first year. |
Why the number can fall short
The view lists one database at a time and leaves out empty tables, so a missing database means missing gigabytes. If it returns no rows at all, the user probably is not a superuser and has not been granted access.
03
What should you know about your Amazon Redshift data before you apply?
The short answer
Have three facts ready: the total size of your tables in GB, how many years of records they hold and how many customers or users the records describe. They go into the data manifest, the written summary of which sources hold your records, how many years each covers and roughly how much. You don't need to run an export to find them.
The size in GB
Write down the total of size in GB. The figure is a count of 1 MB blocks as Redshift stores them, which is after column compression, and it still contains rows flagged for deletion that vacuum has yet to clear, so an uncompressed export will come out at a different size. For live row counts, use estimated_visible_rows and not tbl_rows.
The years of history
The years of history come from the earliest date in your largest fact tables.
The number of seats
The seat count is the number of customers or users whose activity the tables record. The database users who run queries are not seats.
What you don't need to do first
You don't need to export anything to apply, and you don't need to clean or scrub the records. The data vendor anonymizes them before anyone else sees them, and nothing moves until a lawyer-reviewed license is signed. You only need the size, the years and the count for the first conversation. The application, the intro call and each later step are laid out in how to sell your company's data.
04
How do you export your full Amazon Redshift history?
The short answer
A user with an IAM role that can write to an Amazon S3 bucket runs the UNLOAD command once for each table, for example to Parquet files. IAM is the AWS system that decides who may do what, and S3 is Amazon's file storage service. The files are written in parallel, and a manifest file lists every one of them.
The steps with UNLOAD
Each step is run by a developer or database administrator. The command is described in the UNLOAD reference.
- Give the cluster an IAM role that is allowed to write to your S3 bucket. Either associate the role with the cluster or set it as the cluster's default role. The user running the command also needs
SELECTpermission on the tables. - Unload each table to Parquet, a compact column-based file format:
UNLOAD ('SELECT * FROM public.orders') TO 's3://my-bucket/export/orders/' IAM_ROLE default FORMAT AS PARQUET MANIFEST VERBOSE; - If the bucket is in a different AWS Region from the cluster, add
REGION 'us-west-2'to the command. - Check the output in the bucket. Redshift writes at least one file per slice, the unit into which a cluster divides its data and work, and
MANIFEST VERBOSEadds a manifest that lists each file with its row count.
Formats and limits
| Topic | What applies |
|---|---|
| Default format | Pipe-delimited text. CSV, JSON and PARQUET are also supported. |
| File size | Up to 6.2 GB per file by default. MAXFILESIZE can be set anywhere from 5 MB to 6.2 GB. |
| Parquet | Each row group is compressed with Snappy. TIMESTAMPTZ columns lose their time zone. |
| Types that need text or CSV | Columns of type GEOMETRY, HLLSKETCH or VARBYTE can be unloaded only as text or CSV. |
| The outer SELECT | It cannot use LIMIT. |
| Encryption | Output files are encrypted with SSE-S3 (Amazon's default server-side encryption) unless you choose otherwise. |
What an export includes and leaves out
An unload writes the rows that the SELECT returns, which means live rows. Time zones on TIMESTAMPTZ columns are not carried into Parquet files, and the file layout depends on the number of slices in your cluster.
How long an export takes
AWS publishes no time, because it depends on table size, the cluster's size and how busy it is. Because the files are written in parallel across slices, large tables are split into many files, and a quiet period is the kinder choice for other people using the cluster.
Sources: Amazon Redshift documentation, SVV_TABLE_INFO, UNLOAD and Amazon Redshift performance metrics in CloudWatch. Steps checked against these pages in November 2026.
05
What stays private when Amazon Redshift data is licensed?
The short answer
The company decides which tables and columns leave Redshift, and the data vendor replaces names and other personal details with consistent stand-ins before an AI lab sees anything. The company keeps ownership of its data, and the license covers defined records for a defined use.
You choose what is in scope
UNLOAD accepts any SELECT statement, so you can leave out names, emails or card fields and export only the columns you agree to share. This example writes three columns of one table:
UNLOAD ('SELECT order_id, created_at, status FROM public.orders')
TO 's3://my-bucket/export/orders/'
IAM_ROLE default
FORMAT AS PARQUET;The data vendor also deletes records too sensitive to clean safely instead of anonymizing them.
Names are replaced before a lab sees them
Anonymization means replacing every detail that identifies a person or a company, such as a name or an email address, with a stand-in that is the same every time it appears. The records still link to each other and the dates still run in order, so a lab can learn from them, but they no longer say who anyone is. The diagram shows one set of records before and after.
The method, including what is deleted outright and how you review samples, is in how company data is anonymized for AI training.
Ownership and payment
Licensing gives the buyer limited rights to a copy, while the company keeps ownership of the originals. The terms that matter, such as exclusivity, permitted use and deletion, are covered in data licensing. micro1 pays in full within 30 days of accepting the data, usually as a single payment.
Questions
Frequently asked questions
What does the size column in SVV_TABLE_INFO measure?
Blocks of 1 MB of stored data, so 1024 of them make a GB. Divide your total by 1024.
Why does SVV_TABLE_INFO return no rows for me?
Most likely because your user is not a superuser, since the view is visible to superusers only by default. A superuser can run GRANT SELECT ON svv_table_info TO your_user; to open it up, and note that the view also skips empty tables and shows only the database you are connected to.
How large are the files UNLOAD creates?
By default each file can reach 6.2 GB, and Redshift writes them in parallel across slices. If you want smaller files, set MAXFILESIZE anywhere between 5 MB and 6.2 GB.
Why is the SVV_TABLE_INFO size different from the size of my exported files?
Because they measure different things. The view counts compressed blocks and rows marked for deletion, while an unload writes live rows in the format you chose, such as Parquet with Snappy compression.
Can I unload to something other than Amazon S3?
Redshift's UNLOAD command writes to Amazon S3, so files land in a bucket first, and you move them elsewhere with your own tools afterward.
Do we need to export Amazon Redshift before we apply?
No. The application takes about two minutes and shares no data. You only need your total size in GB, your years of history and your number of customers or users for the intro call, and the size query above gives you the first. The full export is needed later, when the records are transferred after you sign the license.
Can we leave out columns that hold sensitive information?
Yes. UNLOAD runs any SELECT statement, so you list only the columns you agree to share.
What is Amazon Redshift data worth to AI labs and data vendors?
It is priced as part of a whole dataset rather than on its own. Companies are paid $100K to $2M or more per dataset, according to micro1, and the price depends on the years of history and how completely the records show each job. What your data is worth explains the factors.
Bottom line
Know what is in Amazon Redshift, then see what it is worth
You don't need to export anything to apply. Apply in two minutes, describe your records on the intro call, and micro1, the data vendor we typically refer companies to, prices the whole dataset and sends an offer 1–2 days later.
Not legal, tax or financial advice. Figures come from micro1's published pages and public reporting as of November 2026.